Physio AI

Inspiration

• 2 in 3 amateur athletes skip rehab because sessions average \$100+ each
  
• Untreated injuries lengthen recovery by >40 % and raise re-injury risk
  
• AI vision + smartphones can put pro-level therapy in every gym bag
  

What it does

• Records movement with phone camera; detects 33 keypoints in real time
  
• Screens for common sports-injury patterns and flags faulty form
  
• Auto-builds a personalized rehab & strength plan reviewed by licensed PTs
  
• Delivers step-by-step video guidance, timers, and rep-count feedback
  
• Tracks progress, pain scores, and ROM; adapts the plan automatically
  
• On-demand chat / video calls with certified therapists when users need human help
  

How we built it

• Frontend: React Native + Expo for iOS & Android parity
  
• AI core: TensorFlow Lite MoveNet + custom CNN for joint-angle scoring
  
• Backend: Node.js (NestJS), PostgreSQL, AWS Lambda for scalable inference
  
• Security: End-to-end encryption, HIPAA-ready data storage on AWS RDS
  
• Continuous integration with GitHub Actions and Expo OTA updates
  

Challenges we ran into

• Achieving stable pose detection in low-light or crowded backgrounds
  
• Tuning models to different body shapes, apparel, and camera angles
  
• Balancing medical accuracy with a gamified, engaging UX
  
• Navigating HIPAA & FDA Software-as-a-Medical-Device (SaMD) guidelines
  
• Collecting enough labeled movement data while protecting user privacy
  

Accomplishments we’re proud of

• MVP demo analyzes squats & lunges with 92 % form-error precision
  
• 100-exercise library curated by three board-certified sports PTs
  
• Pilot with 25 local athletes cut average rehab time by 28 %
  
• Deployed end-to-end pipeline in < 4 weeks during the hackathon
  
• Won “Best Health Tech” at the regional demo day
  

What we learned

• Real-time verbal cues (“knees out”, “neutral spine”) drive higher adherence than silent visuals
  
• Users trust AI more when a human PT co-signs their plan—even asynchronously
  
• Small friction (camera setup < 10 s) is critical; longer flow causes 30 % drop-off
  
• Athletes value progress graphs and milestone badges as much as pain reduction
  

What’s next for Physio AI

• Integrate Watch & IMU sensor data for load and asymmetry detection
  
• Expand exercise set to overhead & rotational sports movements
  
• Partnership pilots with college athletic programs and CrossFit boxes
  
• Submit for FDA Class II SaMD clearance to unlock HSA/FSA coverage
  
• Launch multilingual support (KO, ES, PT) and community leaderboards by Q4 2025

Built With

• ai
• mern